In [1]:

    

%matplotlib inline
path = "data/redux/"
# path = "data/redux/sample/"
from utils import *


    

    




Using TensorFlow backend.

In [2]:

    

batch_size = 8


    

In [3]:

    

# %cd data/redux


    

In [4]:

    

# %cd train


    

In [5]:

    

# %mkdir ../valid


    

In [6]:

    

# g = glob('*.jpg')
# shuf = np.random.permutation(g)
# for i in range(2000): os.rename(shuf[i], '../valid/' + shuf[i])


    

In [7]:

    

# %ls data/redux/valid | wc -l


    

In [8]:

    

# %mkdir ../sample
# %mkdir ../sample/train
# %mkdir ../sample/valid


    

In [9]:

    

# 复制 而不是 之前的移动 注意
from shutil import copyfile


    

In [10]:

    

# g = glob('*.jpg')
# shuf = np.random.permutation(g)
# for i in range(200): copyfile(shuf[i], '../sample/train/' + shuf[i])


    

In [11]:

    

# %ls sample/train | wc -l


    

In [12]:

    

# g = glob('*.jpg')
# shuf = np.random.permutation(g)
# for i in range(50): copyfile(shuf[i], '../sample/valid/' + shuf[i])


    

In [13]:

    

# %ls ../sample/valid | wc -l


    

In [14]:

    

# %mkdir cats
# %mkdir dogs
# %mkdir ../valid/cats
# %mkdir ../valid/dogs
# %mkdir ../sample/valid/cats
# %mkdir ../sample/valid/dogs
# %mkdir ../sample/train/cats
# %mkdir ../sample/train/dogs


    

In [15]:

    

# %mv cat.*.jpg cats/
# %mv dog.*.jpg dogs/


    

In [16]:

    

# %cd ../valid/
# %mv cat.*.jpg cats/
# %mv dog.*.jpg dogs/


    

In [17]:

    

# %cd ../sample/valid/
# %mv cat.*.jpg cats/
# %mv dog.*.jpg dogs/


    

In [18]:

    

# %cd ../train/
# %mv cat.*.jpg cats/
# %mv dog.*.jpg dogs/


    

In [19]:

    

from vgg16 import Vgg16


    

In [20]:

    

vgg = Vgg16()


    

In [21]:

    

batches = vgg.get_batches(path + 'train', batch_size=batch_size)
val_batches = vgg.get_batches(path + 'valid', batch_size=batch_size)


    

    




Found 23000 images belonging to 2 classes.
Found 2000 images belonging to 2 classes.

In [22]:

    

vgg.finetune(batches)


    

In [23]:

    

vgg.fit(batches, val_batches, nb_epoch=1)


    

    




Epoch 1/1
23000/23000 [==============================] - 1108s - loss: 3.4324 - acc: 0.7842 - val_loss: 2.7022 - val_acc: 0.8305

In [25]:

    

# %mkdir data/redux/results


    

In [26]:

    

vgg.model.save_weights(path + 'results/ft1.h5')
# vgg.model.load_weights(path + 'results/ft1.h5')


    

In [27]:

    

vgg.fit(batches, val_batches, nb_epoch=1)


    

    




Epoch 1/1
23000/23000 [==============================] - 1114s - loss: 2.2125 - acc: 0.8616 - val_loss: 1.9457 - val_acc: 0.8780

In [28]:

    

vgg.model.save_weights(path + 'results/ft2.h5')


    

In [29]:

    

vgg.model.optimizer.lr = .01
vgg.fit(batches, val_batches, nb_epoch=1)


    

    




Epoch 1/1
23000/23000 [==============================] - 1111s - loss: 2.1367 - acc: 0.8661 - val_loss: 1.7801 - val_acc: 0.8880

In [30]:

    

vgg.model.save_weights(path + 'results/ft2_1.h5')


    

In [31]:

    

vgg.fit(batches, val_batches, nb_epoch=1)


    

    




Epoch 1/1
23000/23000 [==============================] - 1102s - loss: 1.9863 - acc: 0.8757 - val_loss: 1.9868 - val_acc: 0.8755

In [32]:

    

vgg.model.save_weights(path + 'results/ft2_1.h5')


    

In [33]:

    

# %mkdir data/redux/test1/unknown


    

In [34]:

    

# %mv data/redux/test1/*.jpg data/redux/test1/unknown/
# vgg.model.summary()


    

In [35]:

    

batches, preds = vgg.test(path + 'test1', batch_size = batch_size, class_mode=None)


    

    




None
Found 12500 images belonging to 1 classes.

In [36]:

    

filenames = batches.filenames
filenames[:5]


    

    
Out[36]:





['unknown/4892.jpg',
 'unknown/35.jpg',
 'unknown/3453.jpg',
 'unknown/1869.jpg',
 'unknown/4370.jpg']

In [37]:

    

preds[:5]


    

    
Out[37]:





array([[ 1.,  0.],
       [ 1.,  0.],
       [ 1.,  0.],
       [ 1.,  0.],
       [ 1.,  0.]], dtype=float32)

In [38]:

    

save_array(path+'results/test_preds.dat', preds)
save_array(path+'results/filenames.dat', filenames)


    

In [39]:

    

preds = load_array(path +'results/test_preds.dat')
filenames = load_array(path + 'results/filenames.dat')


    

In [40]:

    

Image.open(path + 'test1/' + filenames[0])


    

    
Out[40]:

In [41]:

    

isdog = np.clip(preds[:,1], 0.05, 0.95)
isdog[:5]


    

    
Out[41]:





array([ 0.05,  0.05,  0.05,  0.05,  0.05], dtype=float32)

In [42]:

    

ids = [int(f[8:f.find('.')]) for f in filenames]
ids[:5]


    

    
Out[42]:





[4892, 35, 3453, 1869, 4370]

In [43]:

    

result = np.stack([ids, isdog], axis=1)
result[:5]


    

    
Out[43]:





array([[  4.8920e+03,   5.0000e-02],
       [  3.5000e+01,   5.0000e-02],
       [  3.4530e+03,   5.0000e-02],
       [  1.8690e+03,   5.0000e-02],
       [  4.3700e+03,   5.0000e-02]])

In [44]:

    

np.savetxt(path + 'sub1.csv', result, fmt='%d,%.5f', header='id,label', comments='')


    

In [45]:

    

from IPython.display import FileLink
FileLink(path + 'sub1.csv')


    

    
Out[45]:




data/redux/sub1.csv

In [ ]:

    

vgg.model.load_weights(path + 'results/ft1.h5')


    

In [ ]:

    

val_batches, probs = vgg.test(path + 'vaalid', batch_size = batch_size)


    

In [ ]:

    

labels = val_batches.classes
filenames = val_batches.filenames


    

In [ ]:

    

probs = probs[:, 0]
probs = np.round(1 - probs)
probs[:8]